Academic Writing Refiner
Overview
Academic papers are judged not only on their scientific merit but also on the quality of their English. Grammatical errors, awkward phrasing, and non-native patterns (commonly called "Chinglish" when originating from Chinese-English interference) can distract reviewers and undermine credibility.
This skill provides a systematic checklist for identifying and correcting common academic English issues. It draws from the academic-writing-guide repository (327+ stars) maintained by SYSUSELab, which catalogs frequent mistakes observed in student and researcher manuscripts across STEM disciplines.
The approach is checklist-driven: rather than relying solely on automated tools, researchers learn to recognize error patterns and self-edit effectively. This skill is especially useful for non-native English speakers preparing manuscripts for international journals and conferences.
Common Error Categories
Grammar and Syntax Errors
| Error Type | Example (Wrong) | Correction |
|---|---|---|
| Article misuse | "We propose a novel the method" | "We propose a novel method" |
| Subject-verb disagreement | "The results shows that..." | "The results show that..." |
| Tense inconsistency | "We train the model and evaluated it" | "We trained the model and evaluated it" |
| Dangling modifier | "Using gradient descent, the loss decreased" | "Using gradient descent, we decreased the loss" |
| Run-on sentence | "The model converges fast it achieves high accuracy" | "The model converges fast and achieves high accuracy" |
Chinglish Patterns
These are interference patterns common when translating from Chinese thought patterns into English:
- Topic-comment structure. Chinese allows "As for X, Y does Z." English prefers "Y does Z to X."
- Wrong: "As for the dataset, we use ImageNet." - Better: "We use ImageNet as our dataset."
- Redundant verbs. Chinese often uses verb-verb compounds that become redundant in English.
- Wrong: "We can be able to achieve..." - Better: "We can achieve..."
- Missing determiners. Chinese has no articles, leading to dropped "the/a/an."
- Wrong: "Model achieves state-of-art result." - Better: "The model achieves a state-of-the-art result."
- Overuse of "respectively."
- Wrong: "The accuracy and F1 are 95% and 0.93 respectively." - Better: "The accuracy is 95% and the F1 score is 0.93."
- "With the development of..." This opening is overused to the point of cliche.
- Better: Lead with the specific problem or finding.
Self-Editing Checklist
Apply this checklist before submitting any manuscript:
Pass 1: Structure and Flow
- Each paragraph has a clear topic sentence.
- Transitions connect paragraphs logically (however, therefore, in contrast).
- No paragraph exceeds 200 words.
- Figures and tables are referenced in order.
Pass 2: Grammar and Mechanics
- All verbs agree with their subjects.
- Tense is consistent within each section (past for Methods/Results, present for general truths).
- Every acronym is defined on first use.
- No comma splices or run-on sentences.
- Articles (a, an, the) are used correctly.
Pass 3: Style and Conciseness
- Remove "very," "really," "quite," "basically" unless essential.
- Replace "in order to" with "to."
- Replace "a large number of" with "many" or give the exact count.
- Convert passive voice to active where possible.
- Eliminate "it is well known that" and similar filler phrases.
Pass 4: Technical Accuracy
- All numbers have units.
- Table column headers are unambiguous.
- Equations are numbered and referenced.
- Statistical claims include confidence intervals or p-values.
Automated Polishing Tools
While manual review is irreplaceable, these tools serve as a useful second pass:
# LanguageTool CLI for grammar checking
java -jar languagetool-commandline.jar -l en-US paper.tex
# Writefull for academic-specific suggestions (VS Code extension)
# Install from VS Code marketplace: "Writefull for LaTeX"
# textlint for rule-based prose linting
npm install -g textlint textlint-rule-no-dead-link textlint-rule-write-good
textlint paper.mdTool Comparison
| Tool | Type | Academic Focus | Free Tier |
|---|---|---|---|
| Grammarly | Cloud | General + academic | Yes (limited) |
| Writefull | Plugin | High (trained on papers) | Yes |
| LanguageTool | Local/Cloud | General | Yes (full) |
| textlint | CLI | Configurable rules | Yes (open source) |
| Trinka | Cloud | High (academic-specific) | Yes (limited) |
Sentence-Level Revision Examples
Before and After
Before: "In recent years, with the rapid development of deep learning, more and more researchers have paid attention to the problem of image classification, which is a very important task in computer vision."
After: "Image classification is a fundamental task in computer vision. Recent advances in deep learning have renewed interest in this problem, with convolutional and transformer architectures achieving human-level accuracy on standard benchmarks."
Before: "We can observe from Table 1 that our method can achieve better performance than baseline methods in terms of all evaluation metrics."
After: "Our method outperforms all baselines across every metric (Table 1)."
Best Practices
- Read published papers in your target venue. Absorb the register and conventions of journals like Nature, IEEE TPAMI, or ACL.
- Have a native speaker review critical sections. At minimum, ask someone to read the Abstract and Introduction.
- Use consistent terminology. If you call it "feature extraction" in Section 2, do not switch to "representation learning" in Section 4 unless you define the distinction.
- Limit sentences to 25 words on average. Long sentences are harder to parse, especially for non-native readers.
- Revise in separate passes. Do not try to fix grammar, structure, and style simultaneously.
References
- academic-writing-guide -- Writing checklist and tutorials (327+ stars)
- Common Errors in Technical Writing -- Henning Schulzrinne, Columbia University
- Writefull -- AI writing assistant trained on published papers
- LanguageTool -- Open-source grammar checker